/srv/irclogs.ubuntu.com/2013/02/24/#ubuntu-devel.txt

xxiaoi can do python setup.py fine on x86, but it always gave me 'illegal instructions' on ppc, am I missing something on python build env?02:41
Chipzzlogout02:45
Chipzz*facepalm* my bad, ignore pls :P02:45
infinityxxiao: If you can sort out what the illegal insn is... Python could be miscompiled (perhaps has AltiVec enabled when it shouldn't)02:51
infinityxxiao: All our buildds support AltiVec, so we don't catch that in production when it slips in. :/02:52
xxiaoinfinity: on this particular chip i actually had altivec02:53
xxiaoyes i'm trying to chase it down02:53
infinityxxiao: Oh, kay.  I knew some of the fsl chips BenC was working with didn't have Altivec, so that was my shot in the dark.02:54
infinityxxiao: Seems unlikely that python is just plain fundamentally broken on PPC, or it would break a huge number of builds in the distro.02:54
xxiaoso far i had no idea on how to debug python setup.py breakage, googling02:55
xxiaoi wrote a simple setup.py and it worked02:55
xxiaoso probably some 'import' caused this trouble02:55
xxiaoi bootstrapped the rootfs from 12.10 ubunut-core for ppc02:55
xxiaothe kernel is 64bit02:56
xxiaothe python setup.py died immediately, i.e. even 'python setup.py help' will not run02:56
xxiaofrom sphinx.setup_command import BuildDoc Illegal instruction03:07
xxiaoseems like the illegal instruction has something to do with BuildDoc and sphinx03:07
xxiaomaybe I'm missing some sphinx pkgs?03:08
xxiaoeverytime i saw documentation related errors from debian/ubuntu it scares me03:11
xxiaonow I can reproduct this easily: put one liner to a python script:03:14
xxiaofrom sphinx.setup_command import BuildDoc03:14
xxiaothen run it, it will illegal instruction03:14
xxiaos/reproduct/reproduce/03:15
xxiaoahh...i could not run sphinx at all, "sphinx-quickstart' will crash on ppc right away03:16
infinityxxiao: Which version of python-sphinx is that?  The one from quantal or raring?03:18
xxiaoquantal03:18
infinityAlthough, hrm.  Can't be sphinx's fault, it's arch:all.03:18
* infinity spins up a PPC machine to test this.03:19
xxiaothanks03:20
infinityxxiao: So, you say that 'sphinx-quickstart' just crashes instantly for you?03:23
infinityxxiao: Works fine on my POWER5 here, on both raring and quantal.03:23
infinityxxiao: http://paste.ubuntu.com/5560511/03:24
xxiaocorrect, it crashes on my t4240 box03:24
infinityOr do I need to fill in some values, does it crash later?03:25
infinityHrm, no, seemed to go all the way through fine.03:26
xxiaohttp://paste.ubuntu.com/5560515/03:26
infinityWell, your kernel seems to be living in the distant past, but that's probably not the issue.03:27
xxiaowe're upgrading 3.0 to 3.8 these days, will math-emulation matter?03:27
infinityPossibly.  BenC probably knows the answer to that better than I do.03:28
infinityI'm really unfamiliar with the fsl cores.03:28
xxiaoit seems t4240's FPU has one special instruction that diffs from power503:28
infinityMy house is full of old Motorola cores, and new IBM ones.03:28
xxiaoi c, will talk with BenC then03:28
xxiaomeanwhile I will try math-emu03:28
xxiaothanks for the help03:29
infinityLet me try this on an old PPC75003:29
xxiaoI assume you have math-emu on?03:30
infinityI have... Whatever our distro kernels have.03:31
xxiaoi'm bulding kernel with math-emu03:34
xxiaohad to do this with yocto, which means 30 minutes to get a clean build, sigh03:34
BenCxxiao: Shouldn't matter03:34
BenCxxiao: You're going to be in Austin next weekend, right?03:34
xxiaoBenC: yes03:34
BenCFlying out there tomorrow03:34
xxiaothat's early03:34
xxiaobut...welcome to Austin03:34
BenCGoing to Dallas on Monday (Servergy office) and coming back to Austin later on in the week03:35
xxiaowhenever i saw doc-related crash, i thought about fpu, which is troublesome for fsl chips03:35
infinityBenC: Want to try to sort out this illegal insn issue xxiao is seeing?  I can't reproduce here, and I don't have the right hardware to dig deeper.03:35
xxiaowe will have weeklies with servergy03:35
BenCjust cheaper to round trip to Austin (and drive to/from Dallas) rather than make a 3-point trip03:35
BenCinfinity, xxiao: Can you give me a repro case?03:36
infinityBenC: Install python-sphinx, and run sphinx-quickstart03:36
infinityBenC: Possibly quantal-specific.03:36
infinityxxiao: Did you try on raring?03:36
xxiaoinfinity: not yet03:36
xxiaoi will try it on raring tomorrow03:37
infinityBenC: Can't reproduce on any of the IBM cores I throw it at.03:38
BenCinfinity: should crash on start?03:38
infinity(Well, I'm still debootstrapping to test on my 750, but I imagine it'll be fine)03:38
xxiaoBenC: or one liner python : from sphinx.setup_command import BuildDoc will trigger it03:38
BenCraring doesn't seem to show it03:38
infinityBenC: Crashes at startup, yeah.03:38
BenCOn e500mc that is03:38
infinityBenC: http://paste.ubuntu.com/5560515/03:38
BenCChecking quantal03:39
infinityBenC: Then again, I'm willing to blame xxiao's sketchy kernel there, too. ;)03:39
BenCxxiao: what does gdb show as the illegal instruction?03:39
xxiaohold on03:39
BenCDoesn't crash on e500mc on quantal03:40
infinityWorks fine on a 15-year-old PPC750 too.03:43
xxiaocrappy t4240 then03:44
xxiaosphinx* is all pythin scripts, don't know how to gdb the instruction03:44
xxiaoit happens when "from sphinx.something import somefunc"03:45
BenCxxiao: gdb python03:45
BenCxxiao: set args -f sphinx-quickstart03:45
xxiaohttp://paste.ubuntu.com/5560553/03:47
infinity*blink(03:47
infinityUnder gdb, I get a SIGILL.03:47
infinityDouble-U Tee Eff.03:47
xxiao__sqrt_finite ()03:47
xxiaoI KNEW it!03:47
xxiaolast time a floating related PHP bug on ppc haunted me for two days03:47
xxiaoi ended up using integer to work around that, instead of compiling all those php stuff from source03:49
infinityxxiao: Is it dying in OPENSSL_cpuid_setup() for you, or elsewhere?03:51
xxiaothe pastebin is all I saw03:52
xxiaowhy is openssl involved here, curious03:52
infinityOh, mine's a false alarm.03:52
infinityOPENSSL_cpuid_setup just can't trap SIGILL when run under gdb.03:53
infinityxxiao: Nevermind me, I'm just being a doofus.03:53
infinityxxiao: Your crash is elsewhere, and more interesting. :P03:53
infinityxxiao: disassemble 0x0fe9aa9403:56
xxiaohttp://paste.ubuntu.com/5560579/04:06
Bluefoxicyffs I need to replace the wall in this room with a big white board.04:07
xxiaoBenC: math-emu seems helped04:16
xxiaoinfinity: with math-emu now it's ok04:17
xxiaois there nodejs for ppc?04:38
BenCxxiao: nope…v8 needs a lot of porting for ppc04:38
infinityYeah. :/04:38
infinityThere's a lot of v8-using stuff out there these days too.04:39
xxiaoBenC: i'm 'recommending" freescale to become openstack member asap04:39
BenCxxiao: We were going to bring up v8 with Freescale this week :)04:39
infinityBenC: I stumbled on a v8/ppc "port" on github a while back, but I suspect it stalled before it ever really got started.04:39
xxiaothe openstack guys mandated nodejs for its dashboard, that kills ppc/mips04:39
BenCinfinity: It's completely empty from > 1 year ago04:39
infinityBenC: I'd love if someone just wrote a *&%! generic C implementation of v8, to be honest.04:39
infinityBenC: Sure, it might be slow, but at least it would work everywhere.04:40
BenCinfinity: amen...04:40
infinityxxiao: There's a MIPS v8 port that actually works, AFAIK, it's just not upstream.04:40
infinityxxiao: But yeah, nothing for PPC.04:40
BenCxxiao: That's fantastic…openstack is right where we want to position CTS-1000 hardware04:40
infinityBenC: If you can talk anyone into putting resources into v8/ppc, that'd be awesome.  If you can talk anyone into generic C, I won't complain. :P04:41
BenCinfinity: I have a feeling I'll be buying lots of jaeger bombs this week…let's see if it gets things moving04:42
* infinity can't help but think this is the sort of thing IBM would have thrown money/people at 10 years ago, but trying to convince them of it today would be hard.04:42
infinityBut perhaps worth asking benh, on the off chance they've been doing something there and just not been vocal.04:42
xxiaook lxc/12.10 is up04:44
xxiaotime to tie it with openstack via cli, as no dashboard can run without nodejs04:44
Bluefoxicyinfinity:  no way, Websphere is infinitely superior to nodejs, and you should be using it instead.  Ask IBM.04:45
infinityBluefoxicy: v8's embedded in a lot of products other than nodejs.  Lacking the port sucks period.04:46
Bluefoxicyinfinity:  Also, MySQL sucks terribly, but if you need something better there's this database called Oracle that the folks who currently control MySQL would be happy to help you deploy...04:46
infinity(And lacking a generic implementation sucks for new arches)04:46
erickLeehi04:47
BenCxxiao: openstack-dashboard works for me04:48
erickLeecan anyone tell me what is the safe package to attempt modifications?04:48
xxiaohttp://paste.ubuntu.com/5560653/04:48
BenCAlthough, I think I used it on x8604:48
xxiaowhat the hack, it takes 2 minutes for lxc to get a login prompt04:48
Bluefoxicyinfinity:  I suffered major stage 12 burnout recently and have come out incredibly, terribly critical and horrible somehow.  I make no excuses or apologies for this, but keep your eyes peeled for cynicism.04:48
xxiaoBenC: which openstack version? they mandated nodejs for folsom04:49
xxiaoif you're doing Essex it might be fine04:49
BenCxxiao: I don't know the versions, but that's from raring04:49
xxiaoBenC: hmm need try that then04:49
xxiaoBenC: i'm building the whole thing from source, maybe I should do deb install instead04:50
BenCxxiao: Might save you some time :)04:50
xxiaobuild it from source is easy to hack in lxc support, which is not officially supported04:50
infinitynova-compute-lxc is packaged.04:51
infinityYou shouldn't need to build from source for that.04:51
xxiaoyeah will try the package install tomorrow04:51
infinityIt's how we currently do openstack on ARM, due to the lack of paravirt solutions there.04:52
xxiaoi spent the whole day to get it fully built from devstack, just missing nodejs04:52
vibhavYou can get away with misaligned access on x86, right?04:52
xxiaonot sure how ubuntu bypassed that04:52
infinityvibhav: Yes, but don't.04:52
* vibhav grumbles 04:53
infinityvibhav: For two reasons.  (a) it's bad practice that creates unportable code and (b) it's slower, cause you waste cycles on fixups.04:53
vibhavinfinity: Indeed. Thoggen had failed to build on arm for this reason04:54
vibhavAnd I was finding ways to fix it04:54
xxiaoBenC: who can I ask further on openstack/raring/ppc?04:56
xxiaoesp the openstack version and nodejs dependency04:56
lifelessxxiao: whats wrong with nodejs ?04:58
BenCxxiao: If there was a nodejs dep, I would not be able to install it :)04:58
BenClifeless: it doesn't run on ppc04:58
BenCxxiao: *wouldn't04:58
BenCerr, had it right I guess04:58
xxiaook will try install from pkgs tomorrow05:03
xxiaohttp://packages.ubuntu.com/quantal/python-django-horizon shows no nodejs05:03
xxiaohttp://changelogs.ubuntu.com/changelogs/pool/main/h/horizon/horizon_2012.2-0ubuntu2/changelog05:04
xxiaoDrop node-less dependency (LP: #1024326)05:04
ubottuLaunchpad bug 1024326 in horizon (Ubuntu Quantal) "node.js is required for access to the dashboard" [High,Fix released] https://launchpad.net/bugs/102432605:04
xxiaothat is it I think05:04
lifelessBenC: ah; shallow or deep issues do you think ?05:05
BenClifeless: deep…v8 needs to be ported05:11
BenCxxiao: right, but it jut means you need nodejs to access it, not to run it, right?05:11
BenCI'm wrong, looks like it needs it on the host system05:12
BenCxxiao: But it does look like it's fixed in quantal and raring05:14
lifelessso the nodejs dep05:14
xxiaoBenC: yes05:14
lifelessis currently just as a compiler for lesscss05:14
lifelessso you can run that once per revision on any architecture and then its done.05:15
xxiaoBenC: basically it tells openstack don't do dynamic compilation, use a pre-built css instead05:15
xxiaoi'm pulling that into devstack now, will see how it goes05:15
lifelesshowever its likely upstream will want to do much more dynamic things later, so I wouldn't bet on it staying as an indefinite workaround05:15
xxiaolifeless: that's why I want to ask freescale to get on openstack paid membership to vote against this tech decision05:16
xxiaoor port v8/nodejs to ppc05:16
lifelesspaid membership gets a board position; nodejs is a tech decision that the board doesn't vote on05:16
xxiaorequires nodejs just for the sake of css seems odd to me05:17
lifelessthe way to get technical leverage is to contribute05:17
lifelesstechnically05:17
lifelessyou could write a python LESS compiler05:17
lifelessthen get the nodejs dep dropped, as long as it has not expanded in scope05:18
=== doko_ is now known as doko
=== zequence_ is now known as zequence
=== highvolt1ge is now known as highvoltage
=== Zilvador_ is now known as Zilvador
jbichahi, could we get a rebuild of cmake, gvfs, and packagekit for the libarchive transition?15:10
Laneyjbicha: what about all the others?19:40
infinityLaney: I'm working on the whole stack right now.19:43
infinityLaney: Though cmake might be stuck on an FTBFS in the testsuite.  *sigh*19:44
jbichainfinity: thanks19:45
* infinity scratches his head at cmake's newfound hatred for its testsuite.19:46
xxiaoI was told turning on math-emu is the wrong thing to do, not sure if that's a correct suggestion20:16
xxiaoinfinity: for ppc of  course, if kernel has math-emu off, will the sphinx-quickstart abort?20:17
xxiaoI would assume ubuntu rootfs is soft-float by default?20:17
xxiaoBenC: pinh20:25
xxiaoping20:25
infinityxxiao: Why would you assume Ubuntu is soft-float?20:25
xxiaoinfinity: just guessing, is it?20:26
infinityIsn't soft-float on PPC a reasonable new (and incompatible) ABI?20:26
xxiaoinfinity: in my case, ppc rootfs in embedded space is mostly soft-float20:26
xxiaoesp those e500 chips20:27
xxiaoif everything is hard float, then why we need math-emu?20:28
infinityI'm not familiar with the math-emu you keep talking about.20:29
infinityIs this specific to fsl kernels?20:29
infinityIf so, it's probably trapping instructions that don't exist on your cores.20:29
xxiaomath-emu is a kernel option that traps floating point instruction if there is no fpu20:40
=== Ursinha_ is now known as Ursinha
=== Ursinha is now known as Ursinha-afk
=== Ursinha-afk is now known as Ursinha
=== Ursinha is now known as Ursinha-afk
=== Ursinha-afk is now known as Ursinha
wooo What is vnode? Is is somewhat similar to inode?23:39
gkcnwooo, http://www.linuxquestions.org/questions/linux-kernel-70/difference-between-inode-and-vnode-657954/23:45

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!